Earlier this week I wrote about the Google DeepMind and Fenris Creations partnership and came away cautiously excited. So I took Friday off from work to make sure I could watch the LIVE broadcast from the Harpa Concert Hall. Then earlier today, Hilmar and DeepMind's Adrian Bolton took the Fanfest stage for 'The Infinite Game' session, and I walked away with considerably less.
Like a lot of those who listened in, I came away fairly empty-handed from that conversation. Not disappointed, but very much hungry for more. As in a lot more. Yes, yes, there was lots of framing, lots of "excitement" about the partnership's potential, about AI lifting the popularity of games like Chess and Go, but very little about what it actually means for the game we play.
So I thought I could give it a try myself and dive into how I'm hoping DeepMind's research could realistically be applied and leveraged in EVE. Basically trying to add the substance that was missing from their talk. Trying to imagine what that near future could be.
I'm probably going to be way off mark, then again maybe not. We'll see.
But I thought I could give it a try.
Here goes nothing...
- NPCs that actually adapt to how you fly — not harder stats, different behaviour
- An AI-powered NPE in Exordium that keeps rookies alive past the two-hour quit window and hands them off to real player run corporations
- Dynamic universe events that respond to what players are doing, not what a designer scripted
- Behind-the-scenes economic modeling that could make balance patches less painful
- The stuff I think people are overhyping — and why TiDi probably isn't going anywhere
It's a long read, but if you care about where New Eden is headed, you'll want to stick around.
So grab a coffee, tea or a cold one, and let's dive in.
Let's fast-forward five years.
It's May 2031.
DeepMind has had half a decade of research access to an offline copy of New Eden. Fenris has been bridging the gap between lab results and Tranquility.
Here's what I think is plausible, what I think is hopeful but unlikely, and what's probably just wishful thinking.
The most likely transformation: NPCs that actually think
This is easily the lowest-of-hanging fruit and the highest-impact change in my humble opinion.
Currently, EVE NPCs — mission agents, belt pirates, even Drifters and Triglavians — follow scripted behaviour trees. They aggro based on rules, orbit at fixed ranges, and use abilities on timers. Old and experienced players read them like an open book. You learn the pattern once, then you optimize around it forever. PvE in EVE has always been something you solve and then farm, not something that challenges you on the hundredth run.
Five years of DeepMind learning EVE PvE combat and mining missions could produce NPC opponents that actually adapt. Not "harder stats on the same script" but genuinely different behaviours. Agents that remember and adapt to the way you complete missions dynamically. Rats that recognize when you're kiting them and change approach. Mission NPCs that focus fire intelligently, call for reinforcements based on what you're flying, and retreat when they're losing. AI NPCs adapting to player behaviours, making sure the game experience is always pushing you outside your comfort.
This would be transformative for PvE, which has not been EVE's strongest content pillar, and the first impressions newbies get of the game. Suddenly ratting isn't just an ISK faucet you optimize and forget. It becomes an evolving challenge. The meta shifts not just because of patch notes, but because the NPCs got smarter.
The Exordium connection
Next is where it gets interesting for the current trajectory. Fenris announced Exordium back in mid-April as the new player experience region. If they're applying DeepMind's learnings to Tranquility by 2031, Exordium could be the first testing ground.
Before you shout back in horror, here me out!
Imagine an NPE where AI-driven mentor agents actually respond to how a new player behaves — not a scripted tutorial that marches you through the same steps regardless, but a system that recognizes "this player is exploring the market mechanics" versus "this player wants to shoot things" and dynamically adjusts what content it pushes your way. That's directly in DeepMind's stated research area of continual learning and long-term planning.
The numbers on why this matters are brutal. At EVE North in 2019 (Toronto, Canada), Fenris (CCP Burger, or should I say FC Burger) shared data showing that nearly 90% of new players who sign up for EVE quit within the first seven days. A 2016 Fanfest keynote was even grimmer: over 1.5 million new players signed up in a single year, and the overwhelming majority permanently quit after just two hours. Executive producer Andie Nordgren's reaction to that stat was two words: "It's super depressing." EVE has never had a problem attracting new players. It has a catastrophic problem keeping them.
An AI system that meets new capsuleers where they are on their journey of learning to play EVE — rather than where a designer assumed they'd be, could be the single biggest retention improvement in the game's history. And Exordium, as a dedicated region for new players, is the obvious place to deploy it without risking the "main" sandbox. Or at the very least to test it.
Now, I'm not naive about how the community may/will react to this.
We've already seen the tension play out. When Fenris launched Aura Guidance earlier this year — an AI chatbot trained on 5.8 million Rookie Help messages — the reaction was exactly the split you'd expect. The studio went out of its way to stress in a devblog that the system "does not give fleet advice, market strategies, or detailed fitting optimization" and "does not replace the depth of knowledge you get from experienced players, mentors, or specialized communities." Fenris was practically tripping over itself to reassure veterans that this wasn't the beginning of the end.
And the community's concern wasn't irrational. As many players pointed out, if newbies stop talking to humans because the bot is "good enough," EVE loses its most powerful retention tool: its community. That organic moment where a veteran drops a million ISK on a confused rookie in Rookie Help chat — that's the hook. That's what makes people stay. Replace it with a chatbot and you've optimized away the very thing that makes EVE special.
The data backs this up. At a 2019 presentation, CCP Hilmar identified what he called "the magic moment" — when a new player loses their ship for the first time, that's when they decide to stay or quit. And the single strongest predictor of which way they go? Whether they belong to a social group in the game. Not better tutorials. Not smoother UI. People.
So the goal of an AI-driven mentor system in Exordium shouldn't be to teach new players EVE. It should be to keep them in the game long enough — past those lethal first two hours, past that first ship loss — to figure out what kind of capsuleer they're becoming, and then hand them off to the real players and corporations that match their interests.
That's the job. Not companion. Not teacher. Matchmaker.
Here's what that could look like. The AI reads that you're gravitating toward exploration and connects you with Signal Cartel, or similar a corporation. It notices you keep picking fights (and loosing them) and nudges you toward a PvP corp, maybe one engaged in Factional Warfare. It sees you spending all your time on the market and introduces you to a trading community. The handoff from artificial guide to human community becomes the graduation moment — the point where EVE stops being a tutorial and starts being EVE.
That model respects the social contract in a way that a permanent AI companion never could. The AI doesn't replace the player-driven experience; it gets new players to the door of one.
And it solves the real problem, which was never "EVE lacks content for new players" but "new players can't find the people who would make them stay."
Dynamic universe events that aren't scripted
Fenris has tried this before with live events and invasion arcs, but they've always been hand-authored with predetermined outcomes or branching paths. Someone at the studio decides "Triglavians invade these systems on this date," and players react to the script.
An AI system trained on EVE's emergent behavior could generate events that respond to what players are actually doing. Not "invasion begins on patch day" but "an AI-driven faction notices a power vacuum in a region where a major alliance just collapsed and moves to exploit it." Content that emerges from the game state rather than being imposed on it.
This is plausible but harder than it sounds. And maybe a tad riskier.
The risk is that AI-generated content feels generic or random rather than meaningful. EVE players care deeply about narrative coherence and political context — they want to know why something is happening, not just that it's happening. Fenris would need to maintain storytelling quality on top of the AI's strategic logic, which is still something humans do better. But the combination — human narrative designers working with AI-driven strategic behavior — could produce something neither could achieve alone.
The invisible revolution: economic modeling
This one is less visible to players but potentially huge.
EVE's economy is notoriously hard to balance. Fenris has economists on staff, but the interactions are so complex that interventions often produce unintended consequences. Nerf a blueprint and three markets you didn't think about crash. Introduce a new resource sink and inflation shifts in ways nobody predicted.
An AI system that has spent five years modeling EVE's economic flows could give Fenris vastly better predictive tools — "if you nerf this blueprint, here's the cascading effect on 47 related markets over the next six months." That doesn't change what players see directly, but it means fewer economy-breaking patches, more confident design decisions, and a healthier game overall.
Players won't notice this one. They'll just notice that balance changes feel less jarring and the economy feels more stable. Which, ironically, might be the most valuable outcome of the entire partnership.
What I'm skeptical about
I don't think Fenris puts DeepMind-trained combat agents into nullsec or lowsec as hostile entities that fight players with player-level tactics. The political backlash would be enormous — "Fenris is using Google's AI to kill our fleets" — and it undermines the core social contract that EVE is a player-driven universe. If it happens at all, it stays contained in PvE content.
I'm also skeptical about any fundamental server architecture changes coming from this partnership. TiDi will probably still exist in 2031. DeepMind is a research lab focused on intelligence, not infrastructure engineering. The server improvements will come from Fenris's own engineering team making incremental progress, not from AI breakthroughs.
And I'd temper expectations about the pace. Five years sounds like a long time, but going from "research paper" to "stable live game system used by tens of thousands of players" is a brutal pipeline. The first two to three years are probably mostly research with small controlled experiments. Meaningful Tranquility integration likely starts showing up around year three or four, with big visible features closer to year five.
If you're expecting revolutionary changes at Fanfest 2027, you're probably a year or two early. Maybe even longer.
The shift nobody's talking about
The most profound change might not be any single feature.
It might be that EVE becomes an MMO where the universe genuinely feels alive — where the space between player interactions isn't dead but populated by systems that learn, adapt, and create pressure.
Not AI that replaces players. AI that fills gaps between them.
That's the version of this partnership that makes EVE not just a better game but a different kind of game, one small AI-optimized mechanic at a time. Whether Fenris can deliver it in five years is the open question. But the pieces are there — the research partner, the financial independence, the revenue momentum, and a game world complex enough to actually stress-test whatever DeepMind builds or learns from the game.
For the first time in a while, I'm looking at EVE's future and seeing something different from what we've seen the last 25 years or so.
That alone is worth paying attention to.
Fly safe. o7
Source articles on new players retention
- EVE Online Developers Discuss Ways To Stop New Players From Leaving The Game - Kotaku
- EVE Online has no problem getting new players, but retaining them is a hell of an issue - Destructoid
- Aura Guidance - Fenris Creations DevBlog
- High Sec and New Player Experience Are Tied - Make a change CCP - EVE Online Forums
- EVE Fanfest 2015: Using Science to Help Newbros - YouTube
- EVE Down Under 2019 - Beyond The Friendship Machine, by Hilmar "CCP Hellmar" Pétursson
No comments:
Post a Comment